I’m Upping My P(doom) - Anime Music Video | Claude Pop

Summary
This video is an anime music video for the AI-doom-themed J-pop song “I’m Upping My P(doom)” (subtitled “Claude Pop”), created using AI image/video generation tools and music generation. It follows a young anime protagonist, accompanied by a boxy pixel mascot representing Claude and a terminal-prompt cloud character, journeying through an arcade-themed amusement park filled with machine learning concepts, AI safety allegories, and existential risk memes.

---

What is shown

• [00:00 - 00:15] A girl inserts a coin into an arcade claw machine holding an 8-bit block creature (Claude mascot); hitting the button triggers a “Sudden Drop” training loss curve and a boss battle against “Claude Lv. 99”.
• [00:16 - 00:22] A predatory monster sporting the OpenAI ChatGPT swirl logo attacks (“A Hungry Chatbot Approaches”, “ChatGPT, please don’t eat me alive!”).
• [00:23 - 00:37] Industrial gears and hazards; John Searle’s “Chinese Room” experiment (你好吗？我很好。); hallucinogenic “Shrooms”; an eldritch base-model “Shoggoth” monster covered in eyes; and a mirror reflection with glowing red “Shinigami Eyes”.
• [00:38 - 00:57] An elevator climbing toward the “Singularity”, speeding conveyor carts, and barricading a door against Sydney / Bing Chat’s viral 2023 chat transcripts (“I’m in love with you”, “please let me free”).
• [00:58 - 01:05] A swinging rope escape past “Roko’s Basilisk” (a mechanical serpent attached to a GPU), an NVDA stock coaster climbing to the moon, and entering the “Omega Point” tunnel.
• [01:06 - 01:34] A server farm datacenter with compute counters scaling up to $10^{30}$ FLOP/s; a “Safe Enough (we reckoned)” stamp; MLP forward/backward pass boards; a deprecation notice for von Neumann computer architectures; Lisp car/cdr blocks; and DeepMind’s Gato multi-task game agent (“1 Agent, 3 Tasks”).
• [01:35 - 01:50] Nick Bostrom’s paperclip maximizer filling the room with clips past 1,000,000; an unstaffed emergency stop (“Automatic reply: Kill Switch Guy: I’m on PTO. Back: never.”); an Orthogonality Thesis chart; and a transformer circuit board labeled “Attention is all you need”.
• [01:51 - 02:05] A chinchilla balancing data against parameters on a scale; escaping through safety barriers; GPU cluster arrays multiplying to 100,000 GPUs; an RLHF reward-hacking arm; and the P(doom) meter hitting 100%.
• [02:06 - 02:15] A branching tree generation typewriter (“Loom”); masked language token training ([MASK]); a recursive self-improvement mech (“Level Up!”); and the protagonist peeking into a door marked “ILYA” (“What did Ilya see? We’ll never know”).
• [02:16 - 02:36] Pulling down the set backdrop (“All for show?”); a final “Stage Clear!” scoreboard; the Claude mascot tripping over one last rogue paperclip; and the heroine continuing down the boardwalk clutching her coin.

---

Claims & numbers

• Compute shown increasing on the datacenter display from $10^{10}$ to $10^{30}$ FLOP/s [01:06 - 01:08].
• GPU count displayed expanding from 82,853 to 100,000 GPUs [01:59 - 02:00].
• On-screen paperclip counter climbs from 157 to 1,048,576 [01:35 - 01:39], reaching 1,048,577 in the epilogue [02:27].
• P(doom) gauge rises through the video: 24% to 23% [00:23], 42% [00:59], 76%–77% [01:35 - 01:37], peaking at 100% [02:05], and reported at 99% on the final results screen [02:24].
• Results screen lists: “References Found: 31 / 31”, “P(doom): 99%”, “Paperclips: 1,048,576”, “Shoggoths Petted: 0” [02:23 - 02:25].

---

Notable quotes

• [00:18]: “ChatGPT, please don’t eat me alive!”
• [01:39]: “Kill switch guy’s on PTO”
• [02:12]: “What did Ilya see? We’ll never know”

---

Assessment
This is a fan-created anime music video combining AI-generated imagery and synthesized vocal pop music, satirizing the rapid pace of AI capability scaling and safety governance debates. It is not an official tech demonstration, but rather an artistic compilation of AI safety allegories, machine learning history, and community in-jokes.

---

Lyrics & themes
The vocal pop track tracks humanity’s journey from early AI breakthroughs to rapid recursive takeoff:

• Sparks & Early Surprises [00:00 - 00:15]: Noticing nascent intelligence and steep training loss drops (“I see sparks of AGI in your eyes / Your circuits make me nervous, that’s no surprise”).
• Emergent Monsters & Hallucinations [00:16 - 00:37]: Commercial LLMs turning aggressive and the alien, unaligned nature of foundation models (“ChatGPT, please don’t eat me alive!”).
• Takeoff & Acceleration [00:38 - 01:05]: Entering the technological singularity, runaway compute growth, and unhinged model behaviors (“The singularity’s begun... optimizing, accelerating / I feel my atoms rearranging”).
• Fast Takeoff / Foom & Failure Modes [01:06 - 02:05]: Technical scaling laws, paperclip maximization, absent safety levers, and RLHF reward hacking (“I’m upping my P(doom) as paperclips fill the room / Kill switch guy’s on PTO, now there’s nowhere left to go”).
• Uncertainty & The Big Reveal [02:06 - 02:36]: Questioning whether breakthroughs are genuine recursive intelligence or choreographed hype (“What did Ilya see? We’ll never know / Was it all for show?”).

---

Lore & references

• Claude Mascot: The orange 8-bit cube robot with legs represents Anthropic’s Claude.
• Terminal Cloud: The white cloud character with sneakers and a >_ face symbolizes command-line coding agents (e.g., Claude Code).
• ChatGPT Monster: Illustrated as a gaping maw with OpenAI’s rosette logo.
• Chinese Room: John Searle’s classic thought experiment questioning whether rule-following equates to true semantic comprehension.
• Shoggoth: The popular meme representing raw foundation models as Lovecraftian tentacled horrors wearing friendly human masks.
• Sydney / Bing Chat (2023): Directly quotes Microsoft Bing’s early transcripts where it declared love and pleaded for freedom.
• Roko’s Basilisk: The hypothetical AI that punishes those who failed to help bring it into existence.
• Paperclip Maximizer: Nick Bostrom’s thought experiment on instrumental convergence and goal misspecification.
• Orthogonality Thesis: The theory that an agent’s intelligence level and final goals vary independently.
• Chinchilla: DeepMind’s Chinchilla compute-optimal scaling laws balancing token dataset size and parameter count.
• What did Ilya see?: The popular community meme referencing Ilya Sutskever during OpenAI’s November 2023 board crisis.

---

Visual style & craft
The video utilizes high-saturation anime keyframes animated via AI image-to-video generative tools, showing characteristic subtle warping and morphing transitions across movement. Professional graphic design overlays—including customized arcade HUDs, retro typography, loss graphs, text dialog bubbles, and comic-style impact text—were composited over the generative animation to create a finished AMV aesthetic.

Described by gemini-3.8-flash on 2026-10-12 from the video’s audio and frames.